Makers’ Marks: Physical Markup for Designing andFabricating Functional Objects

Valkyrie Savage?, Sean Follmer�, Jingyi Li?, Bjorn Hartmann?

?UC Berkeley EECS, �Stanford Universityvalkyrie@eecs.berkeley.edu, sfollmer@stanford.edu, noon@berkeley.edu, bjoern@eecs.berkeley.edu

Figure 1. Example objects designed with the Makers’ Marks system. Left, modeled and annotated object geometry. Center, view of 3D-scanned digitalmodels with generated mounts for functional components. Right, completed objects.

ABSTRACTTo fabricate functional objects, designers create assembliescombining existing parts (e.g., mechanical hinges, electroniccomponents) with custom-designed geometry (e.g., enclo-sures). Modeling complex assemblies is outside the reach ofthe growing number of novice “makers” with access to digitalfabrication tools. We aim to allow makers to design and 3Dprint functional mechanical and electronic assemblies. Basedon a formative exploration, we created Makers’ Marks, a sys-tem based on physically authoring assemblies with sculptingmaterials and annotation stickers. Makers physically sculptthe shape of an object and attach stickers to place existingparts or high-level features (such as parting lines). Our toolextracts the 3D pose of these annotations from a scan of thedesign, then synthesizes the geometry needed to support inte-grating desired parts using a library of clearance and mount-ing constraints. The resulting designs can then be easily 3Dprinted and assembled. Our approach enables easy creationof complex objects such as TUIs, and leverages physical ma-terials for tangible manipulation and understanding scale. Wevalidate our tool through several design examples: a customgame controller, an animated toy figure, a friendly baby mon-itor, and a hinged box with integrated alarm.

Permission to make digital or hard copies of part or all of this work for personal orclassroom use is granted without fee provided that copies are not made or distributedfor profit or commercial advantage and that copies bear this notice and the full citationon the first page. Copyrights for third-party components of this work must be honored.For all other uses, contact the Owner/Author.Copyright is held by the owner/author(s).UIST’15, November 9–11, 2015, Charlotte, North Carolina, USA.ACM 978-1-4503-3779-3/15/11.http://dx.doi.org/10.1145/2807442.2807508

Author KeywordsTangible Design Tools; 3D Printing

ACM Classification KeywordsH.5.2 User Interfaces: Prototyping

INTRODUCTION3D printers have evolved from professional equipment in in-dustrial design to staples in makerspaces; 3D scanners are fol-lowing, with inexpensive phone or webcam-based solutionsincreasingly available. As such tools move out of the work-place and into maker clubs, libraries, museums, and homes,two large questions remain: what objects can people makewith these machines, and how can they design those objects?

Novices, who are new to or uncomfortable with 3D CADmodeling, include hobbyists, who prefer hands-on tangiblemodeling to the abstract software task of CAD; students inprogramming courses, who feel more accomplished with cus-tom electronics cases; and artists, who have strong basesin tangible modeling. Various snap-together kits (e.g., lit-tleBits1) and easy-to-use languages (e.g., Scratch [14]) allownovice users to create functional electronics: we tackle themodeling task to help them design free-form objects integrat-ing such electronics, as well as mechanical features.

A look at Thingiverse, an online 3D printing community2,reveals that novices mostly create static and decorative ob-jects, while experienced designers model functional objects,often as assemblies which include existing, non-printed parts1http://littlebits.cc/2http://thingiverse.com

such as hinges or electronics. One reason for this disparityis designing assemblies with functional components in CADremains complicated: designers must specify not only whereto place functional components, but also how to mount themto allow assembly and ensure functionality.

We take inspiration from prior work on prototyping inter-active objects [10] and 3D modeling via physical annota-tion [17]: our goal is to allow novice users to create printablefunctional objects integrating commercially-available com-ponents. We foreground working entirely with physical me-dia rather than CAD software—our formative work suggestsnovices can express design intent by combining sculpting(larger shapes) and annotation of functionality (finer details).

With our system, Makers’ Marks, users first sculpt the overallshape they would like to print, using clay or other physicalmaterials. They then add annotations with physical stickers,indicating placement for functional components. In our pro-totype, supported parts include components for user interac-tion (e.g., joysticks) as well as mechanical parts (e.g., hinges).

Makers’ Marks captures user-created geometry using a 3Dscanner and replaces annotations with precise 3D geometryfrom a library, allowing the indicated components to be ac-cessible and fastened post-print. By employing physical au-thoring for rough shapes and digital tools for precise assem-bly geometry, we hope to enable easier, quicker creation ofcomplex functional objects (see Figure 1).

M-Marks focuses on a subset of functional objects. Itis restricted to rigid, shelled objects whose surface func-tional components require 3D models and additional clear-ance metadata. Created objects can range from whimsicaland decorative (a waving shark) to precise and functional (agame controller). We contribute (1) an approach for tangiblyauthoring objects with sculpted geometry and physical anno-tations; and (2) a pipeline to process these annotations andadd mounting features to 3D models to assemble printed ob-jects and existing parts.

RELATED WORKOur research draws upon work in tangible and sketch-based3D modeling, and digital fabrication with pre-existing parts.

3D modeling in contextResearchers have have looked to tangible interfaces tolower the barrier of 3D content authoring.Tracked buildingblocks—implemented with integrated electronics [5] or exter-nal computer vision [4]—create digital models correspondingto users’ physical manipulation. Smart measurement toolscan copy real-world measurements to digital model dimen-sions [9, 19]. However, these projects cannot capture com-plex freeform geometry, like the body of a game controller.

Other tools allow users to capture some aspects of existingobjects as input to designs: for example to copy their geom-etry in 2D [2] or 2.5D [3]. These projects align with our vi-sion, but target different types of objects. MixFab [21] uses amixed-reality environment to spatially position CAD modelsaround scanned geometry of existing objects. Similarly, we

target integration of existing and novel parts, but we focus ontangible rather than gesture-based modeling.

Some systems focus on model manipulation through annota-tion: pen annotations on a papercraft model can specify ex-trusions or cuts [17], while sketches on photos of a scene canprovide measurements for correct scaling [8]. We also lever-age annotations, but use them to add higher-level functionalcomponents instead of specifying underlying shape directly.

Fabricating with libraries of pre-existing partsRecently, researchers have explored fabricating interactivefunctionality directly through the use of additive man-ufacturing techniques, e.g., by printing light pipes [22]or acoustically-sensed components [6]. However, mass-manufactured standard parts can be cheaper, more durable, oroffer other performance benefits over printed parts. Accord-ingly, other projects have offered ways to integrate existingparts into novel 3D models. Lau, et al., analyze 3D modelsto automatically place existing fasteners and IKEA compo-nents [7], thus easing assembly. Mueller, et al., accelerate 3Dprinting processes by 3D printing high-detail areas aroundassembled low-resolution LEGO blocks [12]. Villar, et al.,built CAD extensions to design enclosures for .NET Gad-geteer sensors and actuators [18]. With Weichel, they gen-erated laser-cuttable enclosures [20]. Although we use .NETGadgeteer components in our work, allow users to tangiblymodel full 3D objects that integrate them alongside other off-the-shelf electronic and mechanical components.

FORMATIVE DESIGN STUDYWe conducted a formative design study with 7 participants (3experts and 2 women) ages 21-55 recruited from a large techcompany and a university. Participants completed three de-sign tasks using clay, pen and paper, or a combination of thetwo. Participants designed 3 assemblies: an ergonomic gripfor a screwdriver (clay), a travel case for office supplies (pa-per), and a bike mount for a cell phone (combination). Wesaid the designs were to be given to another engineer for it-eration, but that they should convey as much of the design aspossible. We recorded these sessions, asked post-test ques-tions, and later analyzed their sketches and sculpted designs.

We saw a wide variety of solutions to our design tasks, butthe most striking result was the use of annotations. In thesketching condition, we observed call-outs and labels to an-notate design sketches, often indicating features which were

Figure 2. In our formative study, we observed use of annotations in boththe sculpting and sketching conditions. Left: Green material indicateswhere a joint should be placed to connect to another part. Right: Anno-tations indicate where complex parts are located on a simple sketch.

Figure 3. Makers’ Marks has users model (a) and annotate (b) geometry physically. Then, they capture their designs with a 3D scanner (c). M-Marksperforms several computational steps to ensure printability and assembleability of the model (d-f). Finally, users print and assemble their objects (g).

challenging for the designer to draw, like an attachment fea-ture labeled “plastic clips to hold phone in place” (see Figure2 right). In the sculpting condition we saw participants usemultiple materials as annotations, for example paired greenclay dots to indicate connectivity of parts (see Figure 2 left).

Sculpting may have been accessible, but it had some fun-damental limitations. Although useful for expressing roughdesign intent and simple geometry, clay is a challengingmedium for creating precise geometry or complex topologieslike handles. Affixing objects with significant weight to clayis problematic, as the clay can deform and crack. In thesesituations, we noted participants using annotations, whetherphysical or sketched.

Novices employed fewer annotations, and did not go into asmuch depth about joining assemblies (e.g., mounting pointsor placement of screws). Experts had more detailed drawings,and called out more features.

Based on these observations, we developed Makers’ Marks(M-Marks), which blends sculpting and annotations to cre-ate functional objects. Annotations give context and describefunction, while clay defines a rough form. Our users did notannotate directly on clay, as tools for this are lacking: wethus use stickers. M-Marks encapsulates many specifics thatnovices ignored, like mount points and fastener placement.

DESIGNING WITH MAKERS’ MARKSWe introduce a typical scenario in which a novice designeruses Makers’ Marks to make a functional object.

A teenage maker wishes to create a safe box to hide a stash ofher favorite items away from her brother. She wants to makea secret code to open the box: if opened without the code,it sounds an alarm. To form the size she wants, she findstwo cardboard tubes and tapes them together. She molds a“MINE” sign in polymer clay for the front of the box (Fig-

electronic button, joystick, Raspberry Pi, camera, gyroscope,IR rangefinder, servo (2-part), processing unit

mechanical hinge, knob, handleother parting line, hole

Table 1. Components currently supported in Makers’ Marks’s library.

ure 3a), then applies annotation stickers marking locationsfor functional components in the final object.

Our maker places marks for three buttons to enter the code, agyroscope to sense when the box is opened or moved, and aspeaker to play an alarm. She also situates an internal pro-cessing unit, one of the components supported in Table 1.During this design process, she can continually modify herdecorations and the locations of the inputs. Once satisfiedwith the design, she adds one more annotation indicating ahinge (Figure 3b), which also implies a parting line: whenthe object is printed and a hinge is attached, it can be openedalong this line to allow her access to the inside.

She scans the box’s geometry and appearance with a 3D scan-ner (Figure 3c). In the M-Marks UI, she indicates she wantsan alignment lip. M-Marks locates annotations on her object,and each is classified as “button”, “speaker”, etc. (Figure 3d).The system reconstructs the 3D pose of each annotation andhollows out the box geometry; then, it adds mounting pointsto attach components after printing (Figure 3e), ensuring theywill not interfere with each other when physically assembled.The box model is split along the parting line defined by thehinge, and the lip is generated (Figure 3f).

Our maker now prints the processed geometry. As it prints,she has time to program the interactions she wants to sense.M-Marks does not assist with the programming; our maker iswell-versed in Scratch [14]. Once the print finishes, she snapsin the input, speaker and processing modules, connects themwith cables, and screws on the hinge of the box (Figure 3g).

IMPLEMENTATIONThe Makers’ Marks pipeline begins after users model andscan an annotated object. The scan’s result is a full-color 3Dmodel with vertex and triangle information for the mesh, aswell as links to texture files (i.e., photographs), and mappingsof texture files to mesh triangles.

M-Marks uses pre-authored component definitions and has asimple GUI that calls a Python script. Using MeshMixer [16],M-Marks first hollows the object keeping a 1.5mm shell.Then it detects and localizes marks (performed in Matlab andC++); checks for interference, replaces geometry, and gener-

Figure 4. M-Marks performs mark detection and localization, interference checking, geometry replacement, and assembly structure generation.

ates the assembly structure (openSCAD) (see Figure 4). Fi-nally, users fabricate and assemble objects.

Component DefinitionTo work with M-Marks, existing components (listed in Ta-ble 1) must be modeled and added to the part library. Foreach physical component (Figure 5a), we maintain severalmetadata. First, we store sticker designs. Second, we storean approximation of the actual component geometry (Figure5b). Third, we store its clearance information, which includesspace for, e.g., electronic connectors or a joystick’s motion(Figure 5c). Finally, we store mounting geometry (Figure5d)—in this case, bosses and snap-fit arms that will later holdthe joystick PCB in place in the printed model (Figure 5e).

Mark Detection and LocalizationWe detect and localize user-placed annotations, or “marks,”in the 2D texture images associated with a scanned object.We then calculate each mark’s 3D pose (Figure 4 left).

Our mark detection algorithms are implemented in Matlab us-ing SIFT [11] and RANdom SAmple Consensus (RANSAC)[1]. We iterate through our library of marks and all texture im-ages associated with the OBJ file generated by the 3D scan-ner. A mark is discriminated by first finding the SIFT key-points between its reference image and a texture image, andthen by using RANSAC to iteratively take random samples ofthese keypoints to find the best-fit model. RANSAC selectsinliers, which we use for a tighter mark match. For each tex-ture image, we save every detected mark’s corner and centerpoints’ (u, v) coordinates. In C++ we map these 2D texturecoordinates to 3D mesh locations. Finally, Matlab processes

Figure 5. For each physical component (a), we store several metadata:b) approximate component geometry, c) clearance information (here:extends below component for ribbon cable connection), and d) mountinggeometry. We then e) fabricate this geometry (shown from back).

the 3D coordinates to compute the mark’s pose. As a result,we obtain translation and rotation information to accuratelyposition components in the final, processed model.

Mark DesignFor optimal classification, each mark has unique featureswhich distinguish it from others; we use descriptive text, vec-tor art, and component photos. To preserve uniqueness, wecannot use duplicate marks: thus if a user wants two buttonson an object, she places two unique button marks. We differ-entiate within a class with different fonts and vector art, keep-ing the image and size constant. Marks are sized to matchrepresented physical components, aiding in space planning.Our marks are adhesive-backed for easy deployment.

Component PlacementOnce we determine component placement, we perform sev-eral checks. Using clearance geometry, we determine a sur-face offset for components: the internal boards of the elec-tronic components cannot intersect the surface of the model,and must be recessed appropriately. We step backwards alongcomponent normals 1mm at a time until we no longer detectintersection. Then, we determine if any user-placed compo-nents intersect internally. If so, for certain components whoseexact location is not critical (i.e., processing units), we canstep forwards along the normal until the parts no longer in-tersect, then extend the captured geometry with a new hol-low addition large enough to fit the component. Future workcould explore elastic deformation techniques [13] to avoidbox-like protrusions. We create mounting geometry from thecalculated offsets. This consists of both adding fasteners andsubtracting space necessary for clearance (e.g., for a joystickto protrude through the surface) (see Figure 4 right).

We use several fastening techniques to attach functional com-ponents. For electronics we create bosses to register parts in3D space and use tabs mounted on flexure bearings, whichparts snap into, to stay in place. These bosses and tabs arescaled to meet the interior surface of the shell. For mechan-ical components mounted on the exterior of the object, suchas hinges, we print internal geometry to capture a hex nut,allowing a machine screw to fasten the part.

For multi-part components which attach two separateobjects—for example the servo, which is mounted on one ob-ject and moves another object—we use a and b marks. Inplacement, these marks generate unique geometries for inter-facing with the two sides of the component.

Assembly StructuresGenerated geometry can assist in assembly, like parting lines(to cut objects in half) and fasteners (to re-attach the halves).

Some user-specified components define a particular partingline implicitly: e.g., a hinge only works when straddling abreak in the object. In cases where such components are notused, users can specify a parting line using annotations. Plac-ing two parting line annotations on opposing sides of the ob-ject increases precision in estimating line orientation (we takethe average parting plane defined by such annotations).

We employ three methods to re-join the halves of a shelledassembly. The simplest uses adhesive and no additional ge-ometry. We can also generate internal mounting bosses toallow for repeatable access to the interior of the objects (seeFigure 6a): to do so we create a uniform field of bosses run-ning perpendicular to the calculated parting plane, then re-move bosses from this field that intersect with user-placedcomponents and intersect the pruned field with the target ge-ometry. Third, we can create a “lip” to assist in alignment,generated using the 2D cross-section of the object’s partingplane. We vertically extrude this outline, offsetting the inneredge to create a 1.5mm thick “lip” (see Figure 6b).

Fabrication and AssemblyMakers’ Marks produces hollow component pieces: usersprint these, remove support material, mount components, andfasten separated parts with hinges or bosses.

M-Marks models have potentially-complex overhangs due toboth shelling and inclusion of mounting geometry at arbitraryangles. Thus, fabrication using a 3D printer that can lay downa removable support material is preferable. While most suchmachines are still limited to the professional market, maker-class printers with multiple extruders are already available,e.g., the Makerbot Replicator 2X. Machines using alternativetechnologies, like sintering powders or curing liquids, mayalso be compatible with the generated shapes.

EXAMPLE OBJECTSTo validate our tool, we created several functional, interactiveobjects using M-Marks. We scanned these on a NextEngine3D scanner and printed them on a Stratasys uPrint SE Plus.

Figure 6. We create bosses (a) that avoid components. These bossescan be fastened together with threaded inserts and screws. We can alsocreate “lips” (b) that help with alignment along a parting line.

Safe BoxThe box, described in Designing with Makers’ Marks, wasmodeled with cardboard tubes wrapped in paper and cus-tomized with clay, showing the variety of materials that canbe used to design with M-Marks. The box combines mechan-ical and electronic features seamlessly — users simply printout different sticker marks. The speaker component’s aes-thetic feature, the grill pattern which cuts through the shell,implies that beyond functional components M-Marks can addpurely aesthetic components to assemblies.

Video Game ControllerMost game controllers are not designed for a specific userand can be too big, too small, or just uncomfortable.We mod-eled a clay controller to fit one author’s hands perfectly andprocessed it using M-Marks: dual joysticks and buttons cancontrol video games. Users can quickly iterate during the pre-scan clay phase, as tangible manipulation has a low barrier.

Animated Left SharkWe sculpted a bust of Left Shark3 to make an animated smarttoy. We annotated its arm and bust with a two part servosticker and its chest with an IR rangefinder sticker. Afterprinting and assembly, Left Shark waves when you walkclose. This example highlights using two-part marks to po-sition two pieces relative to each other that can be sculptedindependently, i.e., the shark body and fin.

Friendly Baby MonitorWe created a friendly baby monitor from a popular children’stoy shaped like a potato. It has a Raspberry Pi and camera in-side, and holes punched through the surface to allow cribsideattachment and WiFi connectivity. We sculpted additionalnose space to fit the camera, but were able to use the rest ofthe body as-is, indicating that even makers without sculptingexpertise can create interesting and unique functional objectswith the M-Marks system.

LIMITATIONS AND FUTURE WORKMakers’ Marks has several important limitations. First, 3Dscanning only captures surface geometry. Internal geome-try is invisible to the scanner—instead, we rely on simpleshelling to create objects with space for mounting additionalcomponents. While some of our library parts, e.g., the RPi,sit entirely inside a printed object, their location has to be in-dicated on the outside. 3D scanning also requires some userexpertise for cleaning meshes: this hurdle is outside our scopeand will likely improve with scanning software.

Adhering stickers to an object’s surface has limited precisionas stickers do not fully conform to underlying object geom-etry. Thus, indicating precise orientation, like a parting linerunning orthogonal to a surface, can be challenging.

M-Marks requires a parts library. Some suppliers provide 3Dmodels of their parts: these only need additional clearanceand mounting metadata. This currently requires measurementand CAD expertise, but is only needed once per part. Future3http://knowyourmeme.com/memes/super-bowl-xlix-halftime-shark

work could explore enabling broader cross-sections of usersto add and contribute to such part libraries, or devise algo-rithms to synthesize mounting geometry automatically.

Our vocabulary of marks can also be extended. We currentlyfocus exclusively on component and parting marks for inte-grating off-the-shelf components. Future work could investi-gate marks that express other attributes—such as surface tex-ture. Another potential avenue could explore combining M-Marks with other sensing techniques for fabricated objects,e.g., the computer vision-based techniques of Sauron [15] orthe acoustic techniques of Acoustruments [6].

CONCLUSIONWe presented Makers’ Marks, a tool which processes physi-cal markup of real-world sculpted designs, adding mountingpoints for mechanical and electronic components and embed-ding knowledge of the fabrication process. Makers’ Marksis informed by the results of a formative design study with3D design novices. We additionally demonstrated a series ofdesign examples created using Makers’ Marks.

ACKNOWLEDGMENTSThis material is based on work supported by the NSF un-der Grant No. DGE 1106400, a Sloan Fellowship, and theCITRIS Connected Communities Initiative. The authors alsothank Evan Savage and Eldon Schoop.

REFERENCES1. Fischler, M. A., and Bolles, R. C. Random sample

consensus: A paradigm for model fitting withapplications to image analysis and automatedcartography. CACM 24, 6 (June 1981), 381–395.

2. Follmer, S., Carr, D., Lovell, E., and Ishii, H. CopyCAD:remixing physical objects with copy and paste from thereal world. In UIST ’10 Adjunct, 381–382.

3. Follmer, S., and Ishii, H. KidCAD: digitally remixingtoys through tangible tools. In CHI ’12, 2401–2410.

4. Gupta, A., Fox, D., Curless, B., and Cohen, M.Duplotrack: a real-time system for authoring andguiding duplo block assembly. In UIST’12, 389–402.

5. Jacobson, A., Panozzo, D., Glauser, O., Pradalier, C.,Hilliges, O., and Sorkine-Hornung, O. Tangible andmodular input device for character articulation. InSIGGRAPH ’14 Emerging Technologies, 24.

6. Laput, G., Brockmeyer, E., Hudson, S. E., and Harrison,C. Acoustruments: Passive, acoustically-driven,interactive controls for handheld devices. In CHI ’15,2161–2170.

7. Lau, M., Ohgawara, A., Mitani, J., and Igarashi, T.Converting 3D furniture models to fabricatable parts andconnectors. In ACM Trans. Graph., vol. 30, ACM(2011), 85.

8. Lau, M., Saul, G., Mitani, J., and Igarashi, T.Modeling-in-context: user design of complementary

objects with a single photo. In Sketch-Based Interfacesand Modeling Symposium ’10, 17–24.

9. Lee, J., Su, V., Ren, S., and Ishii, H. HandSCAPE: avectorizing tape measure for on-site measuringapplications. In CHI ’00, 137–144.

10. Lee, J. C., Avrahami, D., Hudson, S. E., Forlizzi, J.,Dietz, P. H., and Leigh, D. The Calder toolkit: Wiredand wireless components for rapidly prototypinginteractive devices. In DIS ’04, 167–175.

11. Lowe, D. G. Method and apparatus for identifying scaleinvariant features in an image and use of same forlocating an object in an image. US Patent 6,711,293, 032004.

12. Mueller, S., Mohr, T., Guenther, K., Frohnhofen, J., andBaudisch, P. faBrickation: fast 3D printing of functionalobjects by integrating construction kit building blocks.In CHI ’14, 3827–3834.

13. Prevost, R., Whiting, E., Lefebvre, S., andSorkine-Hornung, O. Make it stand: Balancing shapesfor 3D fabrication. ACM Trans. Graph. 32, 4 (July2013), 81:1–81:10.

14. Resnick, M., Maloney, J., Monroy-Hernandez, A., Rusk,N., Eastmond, E., Brennan, K., Millner, A., Rosenbaum,E., Silver, J., Silverman, B., et al. Scratch: programmingfor all. CACM 52, 11 (2009), 60–67.

15. Savage, V., Chang, C., and Hartmann, B. Sauron:Embedded single-camera sensing of printed physicaluser interfaces. In UIST ’13, 447–456.

16. Schmidt, R., and Singh, K. Meshmixer: an interface forrapid mesh composition. In ACM SIGGRAPH 2010Talks, ACM (2010), 6.

17. Song, H., Guimbretiere, F., Hu, C., and Lipson, H.ModelCraft: capturing freehand annotations and edits onphysical 3D models. In UIST ’06, 13–22.

18. Villar, N., Scott, J., Hodges, S., Hammil, K., and Miller,C. .NET Gadgeteer: A platform for custom devices. InPervasive 2012, Lecture Notes in Computer Science(2012).

19. Weichel, C., Alexander, J., Karnik, A., and Gellersen, H.SPATA: Spatio-tangible tools for fabrication-awaredesign. In TEI ’15, 189–196.

20. Weichel, C., Lau, M., and Gellersen, H. Enclosed: acomponent-centric interface for designing prototypeenclosures. In TEI ’13, 215–218.

21. Weichel, C., Lau, M., Kim, D., Villar, N., and Gellersen,H. W. MixFab: a mixed-reality environment for personalfabrication. In CHI ’14, 3855–3864.

22. Willis, K., Brockmeyer, E., Hudson, S., and Poupyrev, I.Printed optics: 3D printing of embedded opticalelements for interactive devices. In UIST ’12, 589–598.